Learning Bipedal Walking for Humanoids with Current Feedback